Atmospheric correction and oceanic constituents retrieval, with a neuro-variational method

Abstract

Ocean color sensors on board satellite measure the solar radiation reflected by the ocean and the atmosphere. This information, denoted reflectance, is affected for 90% by air molecules and aerosols in the atmosphere and for only 10% by water molecules and phytoplankton cells in the ocean. Our method focuses on the chlorophyll-a concentration (chl-a) retrieval, which is commonly used as a proxy for phytoplankton concentration. Our algorithm, denoted NeuroVaria, computes relevant atmospheric (Angstrom coefficient, optical thickness, single-scattering albedo) and oceanic parameters (chl-a, oceanic particulate scattering) by minimizing the difference over the whole spectrum (visible + near infrared) between the observed reflectance and the reflectance computed from artificial neural networks that have been learned with a radiative transfer model. NeuroVaria has been applied to SeaWiFS (sea-viewing wide field-of-view sensor) imagery in the Mediterranean sea. A comparison with in-situ measurements of the water-leaving reflectance shows that NeuroVaria enables to better reconstruct this component at 443 nm than the standard SeaWiFS processing. This leads to an improvement of the retrieval of the chl-a for the oligotrophic sea. This result is generalized to the entire Mediterranean sea through weekly maps of chl-a.